CN116307078A

CN116307078A - Account label prediction method and device, storage medium and electronic equipment

Info

Publication number: CN116307078A
Application number: CN202310053676.6A
Authority: CN
Inventors: 欧阳逸
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2023-02-03
Filing date: 2023-02-03
Publication date: 2023-06-23

Abstract

The embodiment of the application discloses an account label prediction method, an account label prediction device, a storage medium and electronic equipment. According to the method, relationship modeling of each account is conducted according to the behaviors of each account, and therefore a social network diagram is obtained. And carrying out label prediction on each node of the social network graph to obtain label distribution corresponding to each node. Based on the nodes with known labels in the social network diagram and label distribution corresponding to each node, the prediction labels corresponding to the nodes without labels can be predicted in a label propagation mode. In the method, the node information of each node and the label in the node with the known label are utilized in the process of obtaining the prediction label, and the information of the node and the information of the known label are effectively modeled, so that the accuracy of label prediction is improved.

Description

Account label prediction method and device, storage medium and electronic equipment

Technical Field

The embodiment of the application relates to the field of artificial intelligence, in particular to an account label prediction method, an account label prediction device, a storage medium and electronic equipment.

Background

In the related art, account labels are sometimes required to be predicted, and the account labels represent whether certain actions exist in an account. In general, only a part of accounts carry account labels, but more accounts do not carry account labels, and the step of setting the labels on the accounts generally needs to consume a great deal of labor cost and time cost, so that the coverage rate of the account labels is insufficient, and the coverage rate of the account labels further affects the community discovery or the mining effect of the group, so that the mining of group behaviors is also affected. It can be seen that prediction of account labels is important for improving account label coverage and implementing partner mining, but the related technology is not satisfactory in terms of the performance of prediction accuracy of account labels.

Disclosure of Invention

In order to solve at least one technical problem, embodiments of the present application provide a method, an apparatus, a storage medium, and an electronic device for predicting an account label, so as to solve a technical problem of insufficient accuracy of predicting an account label in a related technology.

In one aspect, an embodiment of the present application provides an account tag prediction method, where the method includes:

Acquiring a plurality of accounts related to target behaviors and target behavior execution information corresponding to each account;

extracting preset behavior co-occurrence conditions according to target behavior execution information corresponding to each account, and generating a social network diagram according to an extraction result, wherein the preset behavior is a target behavior meeting preset requirements, and accounts corresponding to two nodes connected by edges in the social network diagram have the same preset behavior;

node classification is carried out on the social network graph to obtain a first type node carrying a label and a second type node not carrying the label, and the label represents whether a corresponding account has target social behavior or not;

performing label prediction on the nodes according to the information of each node in the social network diagram to obtain label distribution corresponding to each node;

predicting the label of each second class node according to the labels carried by the first class nodes and the label distribution corresponding to the nodes.

In another aspect, an embodiment of the present application provides an account tag prediction apparatus, including:

the information acquisition module is used for acquiring a plurality of accounts related to target behaviors and target behavior execution information corresponding to each account;

The label prediction module is used for extracting preset behavior co-occurrence conditions according to target behavior execution information corresponding to each account, generating a social network diagram according to an extraction result, wherein the preset behavior is a target behavior meeting preset requirements, and the accounts corresponding to two nodes connected by edges in the social network diagram have the same preset behavior; node classification is carried out on the social network graph to obtain a first type node carrying a label and a second type node not carrying the label, and the label represents whether a corresponding account has target social behavior or not; performing label prediction on the nodes according to the information of each node in the social network diagram to obtain label distribution corresponding to each node; predicting the label of each second class node according to the labels carried by the first class nodes and the label distribution corresponding to the nodes.

In another aspect, embodiments of the present application provide a computer readable storage medium having at least one instruction or at least one program stored therein, where the at least one instruction or at least one program is loaded and executed by a processor to implement an account label prediction method as described above.

In another aspect, an embodiment of the present application provides an electronic device including at least one processor, and a memory communicatively coupled to the at least one processor; the memory stores instructions executable by the at least one processor, and the at least one processor implements the account tag prediction method by executing the instructions stored by the memory.

In another aspect, embodiments of the present application provide a computer program product comprising a computer program or instructions which, when executed by a processor, implement an account label prediction method as described above.

The embodiment of the application provides an account label prediction method, an account label prediction device, a storage medium and electronic equipment. According to the method, relationship modeling of each account is conducted according to the behaviors of each account, and therefore a social network diagram is obtained. And carrying out label prediction on each node of the social network graph to obtain label distribution corresponding to each node. Based on the nodes with known labels in the social network diagram and label distribution corresponding to each node, the prediction labels corresponding to the nodes without labels can be predicted in a label propagation mode. In the method, the node information of each node and the label in the node with the known label are utilized in the process of obtaining the prediction label, and the information of the node and the information of the known label are effectively modeled, so that the accuracy of label prediction is improved.

Drawings

In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present application or of the related art, the following description will briefly explain the drawings required to be used in the embodiments or the related art, and it is apparent that the drawings in the following description are only some embodiments of the present application, and other drawings may be obtained according to these drawings without inventive effort to those skilled in the art.

Fig. 1 is a schematic diagram of an implementation framework of an account tag prediction method provided in an embodiment of the present application;

fig. 2 is a flow chart of an account label prediction method provided in an embodiment of the present application;

FIG. 3 is a schematic diagram of partner mining through account prediction provided by an embodiment of the present application;

fig. 4 is a flowchart of a training method of an account tag prediction model according to an embodiment of the present application;

FIG. 5 is a flowchart of another training method for an account tag prediction model according to an embodiment of the present application;

FIG. 6 is a schematic diagram of an implementation of the training phase provided by embodiments of the present application;

FIG. 7 is a schematic diagram of an implementation of an application phase provided by an embodiment of the present application;

FIG. 8 is a block diagram of an account label prediction apparatus provided in an embodiment of the present application;

Fig. 9 is a schematic hardware structure of an apparatus for implementing the method provided in the embodiment of the present application.

Detailed Description

The following description of the technical solutions in the embodiments of the present application will be made clearly and completely with reference to the accompanying drawings in the embodiments of the present application, and it is apparent that the described embodiments are only some embodiments of the present application, but not all embodiments. All other embodiments, which can be made by one of ordinary skill in the art without undue burden from the embodiments of the present application, are within the scope of the embodiments of the present application.

It should be noted that the terms "first," "second," and the like in the description and the claims of the embodiments of the present application and the above figures are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that embodiments of the present application described herein may be implemented in sequences other than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or server that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed or inherent to such process, method, article, or apparatus, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.

In order to make the objects, technical solutions and advantages disclosed in the embodiments of the present application more apparent, the embodiments of the present application will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the present application embodiments and are not intended to limit the present application embodiments.

The terms "first" and "second" are used below for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include one or more such feature. In the description of the present embodiment, unless otherwise specified, the meaning of "plurality" is two or more. In order to facilitate understanding of the technical solutions and the technical effects produced by the embodiments of the present application, the embodiments of the present application first explain related terms:

cloud technology (Cloud technology): the hosting technology is used for integrating hardware, software, network and other series resources in a wide area network or a local area network to realize calculation, storage, processing and sharing of data. The cloud computing business model application-based network technology, information technology, integration technology, management platform technology, application technology and the like can be collectively called to form a resource pool, and the resource pool is flexible and convenient as required. Cloud computing technology will become an important support. Background services of technical networking systems require a large amount of computing, storage resources, such as video websites, picture-like websites, and more portals. Along with the high development and application of the internet industry, each resource possibly has an own identification mark in the future, the identification mark needs to be transmitted to a background system for logic processing, data with different levels can be processed separately, and various industry data needs strong system rear shield support and can be realized only through cloud computing.

Artificial intelligence (Artificial Intelligence, AI) is the theory, method, technique and application system that uses a digital computer or a machine controlled by a digital computer to simulate, extend and extend human intelligence, sense the environment, acquire knowledge and use the knowledge to obtain optimal results. In other words, artificial intelligence is an integrated technology of computer science that attempts to understand the essence of intelligence and to produce a new intelligent machine that can react in a similar way to human intelligence. Artificial intelligence, i.e. research on design principles and implementation methods of various intelligent machines, enables the machines to have functions of sensing, reasoning and decision.

The artificial intelligence technology is a comprehensive subject, and relates to the technology with wide fields, namely the technology with a hardware level and the technology with a software level. Artificial intelligence infrastructure generally includes technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/switching systems, mechatronics, and the like. Artificial intelligence software technology includes computer vision technology, speech processing technology, natural language processing technology, machine learning/deep learning, and other major directions.

Machine Learning (ML) is a multi-domain interdisciplinary, involving multiple disciplines such as probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory, etc. It is specially studied how a computer simulates or implements learning behavior of a human to acquire new knowledge or skills, and reorganizes existing knowledge structures to continuously improve own performance. Machine learning is the core of artificial intelligence, a fundamental approach to letting computers have intelligence, which is applied throughout various areas of artificial intelligence. Machine learning and deep learning typically include techniques such as artificial neural networks, confidence networks, reinforcement learning, transfer learning, induction learning, teaching learning, and the like.

Deep learning: the deep learning concept is derived from the study of artificial neural networks, which belongs to the field of machine learning. The multi-layer sensor with multiple hidden layers is a deep learning structure. Deep learning forms more abstract high-level representation attribute categories or features by combining low-level features to discover distributed feature representations of data.

Graph neural network: the graph neural network is a method for graph domain information in deep learning, and in order to simultaneously process side relations with various categories in the graph, the relational graph neural network independently processes the side relations under various categories first, and then aggregates aggregation results of all the categories. The graph neural network is a neural network which utilizes a deep neural network (such as a convolutional network, a cyclic neural network, a self-encoder and the like) to process graph structure data, learn information such as node representation, graph representation and the like, and further complete tasks such as node classification, graph classification, link prediction and the like. Common graph neural networks are: a graph convolution network, a graph annotation network, a graph self-encoder, and the like. For example, the transducers and ViT below may be configured separately or may participate in the formation of the neural network.

Transformer: a model for extracting the mutual relation between every two elements of a sequence by adopting a self-attention (SelfAttention) structure is widely applied to the fields of natural language processing, image processing and time sequence prediction. The transducer is a model based on a multi-head attention mechanism, essentially an Encoder-Decoder model. The Transformer Encoder model is input by word embedded representation of a sentence and its corresponding position-coded information, and the core layer of the model is a multi-headed attention mechanism. The multi-head attention mechanism is to use a plurality of attention mechanisms to perform independent calculation so as to acquire more layers of semantic information, and then splice and combine the results acquired by the attention mechanisms to obtain a final result. The Add & Norm layer sums and normalizes the input and output of the Multi-Head Attention layer, then transmits the sum to the Feed Forward layer, and finally performs the Add & Norm processing again to output the final word vector matrix. The transducer is a full join (or one-dimensional convolution) plus Attention combination. The algorithm has good parallelism and accords with the current hardware environment.

ViT the image classification model is composed of a multi-layer transducer structure, the image is firstly segmented into a plurality of local blocks, and then interaction features among different blocks are extracted through a SelfAttention mechanism, so that feature relations of different positions in the image can be well modeled.

Tag propagation: the label propagation algorithm is to predict labels of unlabeled nodes in the graph by using labels of labeled nodes in the graph according to relations among nodes in the graph and label information of the nodes. Community discovery (Community Detection) is a popular and broad topic, which is actually evolved from the problem of sub-graph segmentation, in a real social network, some users are closely connected, some users are sparsely connected, and a closely connected user group can be regarded as a community, and in a wind control problem, the community can be simply understood as group mining. The current community discovery problems fall into two main categories: non-overlapping community discovery and overlapping community discovery. The non-overlapping community discovery problem is described as: in a network, each node can only belong to the same community, which means that there is no intersection between communities. In non-overlapping community discovery algorithms, there are different kinds of solutions: a modularity-based community discovery algorithm and a tag propagation-based community discovery algorithm. The basic idea of the community discovery algorithm based on label propagation is to update label information of unlabeled nodes by label information of labeled nodes, propagate the label information in the whole social network until convergence, wherein the most representative is the label propagation algorithm (LPA, label Propagation Algorithm). Label Propagation Algorithm, also known as the tag propagation algorithm (LPA), is an algorithm that discovers communities quickly in the graph. In the LPA algorithm, the label of a node is completely determined by its immediate neighbors. The label propagation algorithm is a local community discovery algorithm based on label propagation, and the basic idea is that labels of nodes depend on label information of neighbor nodes, the influence degree is determined by node similarity, and the stability is achieved through propagation iteration updating. In LPA, the algorithm can repeatedly community the label of a node to the label that appears most frequently in the nodes' neighbors after initializing each node with a unique label. The algorithm may stop when the label of each node appears most frequently in its neighbors. The algorithm may be asynchronous in that each node may update without waiting for the rest of the nodes to update.

Group spreading excavation: in case a small number of monomer tags or partner tags are known, tags of other monomers are predicted, and partners are discovered and mined, more partners are detected.

In the related art, the group partner diffusion mining can be regarded as a graph-based semi-supervised learning problem, and the methods commonly used at present can be divided into two types:

the first type is a label propagation algorithm, wherein label propagation is based on node labels and relationships among nodes in a social network graph, and labels of marked nodes are utilized to predict labels of unmarked nodes. A social network graph may be understood as a graph that characterizes relationships between nodes, as determined from their behavior and their own attributes. In the label propagation process, each node updates its own label according to the label of the adjacent node, and the larger the similarity between the label and the adjacent node is, the larger the influence weight of the adjacent node on the label is, so that the labels of the similar nodes tend to be consistent. The disadvantage of this type of method is that only the label information of the nodes in the graph and the connection relationship between the nodes are considered, and the full utilization of the node characteristics of the nodes themselves is ignored.

The second type is a graph neural network algorithm, and the graph neural network aggregates node characteristics through information transmission among nodes in a social network graph so as to predict node labels. The disadvantage of this type of method is that only the characteristic information of the nodes in the graph and the connection relationship between the nodes are considered, and the full utilization of the label information of the nodes themselves is ignored.

In the process of performing group excavation, the embodiment of the application proposes that feature information and label information exist in the nodes in the social network graph formed by the related technology, a label propagation algorithm focuses on considering the label information, and a graph neural network algorithm focuses on considering the feature information, is an information processing mode with emphasis or choice, and cannot comprehensively use the complete information of marked nodes in the social network graph to predict labels of unmarked nodes. Therefore, the embodiment of the application effectively utilizes the characteristic information and the label information of each node in the social network diagram by providing the account label prediction method, and improves the label prediction accuracy of the nodes without the label information. Specifically, the method models the relationship of each account according to the behavior of each account, so as to obtain a social network diagram. And carrying out feature transfer on the node features of each node in the social network diagram, namely learning the node representation of each node according to the transfer result, and mapping the node representation into label distribution corresponding to each node respectively. Based on the nodes with known labels in the social network diagram and label distribution corresponding to each node, the prediction labels corresponding to the nodes without labels can be predicted in a label propagation mode. In the method, the node characteristics of each node and the labels in the nodes of the known labels are utilized in the process of obtaining the prediction labels, and the characteristics and the labels are effectively modeled, so that the accuracy of label prediction is improved. After the labels of the nodes are obtained through prediction, the community discovery algorithm is utilized to conduct community division on accounts respectively represented by the nodes, at least one cluster is obtained, and the cluster in the embodiment of the application can represent the results of the group mining.

Referring to fig. 1, fig. 1 is a schematic diagram of an implementation framework of an account tag prediction method according to an embodiment of the present disclosure, where, as shown in fig. 1, the implementation framework may at least include a client 10 and a server 20, where the client 10 and the server 20 communicate through a network 30, and the implementation framework may also be considered as an account tag prediction system, and the account tag prediction system is used to train an account tag prediction model and provide an account tag prediction service based on the model. The server 20 may be located in a cloud environment, the server 20 predicting nodes in the system for account labels in the cloud environment, the nodes being any nodes in the cloud environment.

The server 20 may first train an account tag predictive model. In the case of obtaining the account tag prediction model, the server 20 may provide the account tag prediction service to the outside. Under the condition that an account label prediction request sent by the client 10 is obtained, a plurality of accounts related to target behaviors and target behavior execution information corresponding to each account are obtained; extracting preset behavior co-occurrence conditions according to target behavior execution information corresponding to each account, generating a social network diagram according to an extraction result, wherein the preset behavior is a target behavior meeting preset requirements, and the accounts corresponding to two nodes connected by edges in the social network diagram have the same preset behavior; node classification is carried out on the social network graph to obtain a first type node carrying a label and a second type node not carrying the label, and the label represents whether a corresponding account has target social behavior or not; carrying out label prediction on the nodes according to the information of each node in the social network diagram to obtain label distribution corresponding to each node; predicting the label of each second class node according to the labels carried by the first class nodes and the label distribution corresponding to the nodes.

The above-described framework of embodiments of the present application may provide account tag prediction capabilities required for applications in various scenarios, including, but not limited to, cloud technology, cloud gaming, cloud rendering, artificial intelligence, intelligent transportation, assisted driving, video media, smart communities, instant messaging, and the like. The components in the framework may be terminal devices or servers. Terminal devices include, but are not limited to, cell phones, computers, intelligent voice switching devices, intelligent home appliances, vehicle terminals, and the like.

Referring now to the method for predicting an account label according to the embodiments of the present application, fig. 2 is a schematic flow chart of a method for predicting an account label according to the embodiments of the present application, where the method for predicting an account label may be performed based on the foregoing server 20. Embodiments of the present application provide method operational steps as described above in the embodiments or flowcharts, but may include more or fewer operational steps based on conventional or non-inventive labor. The order of steps recited in the embodiments is merely one way of performing the order of steps and does not represent a unique order of execution. When implemented in a system, terminal device or server product in practice, the methods illustrated in the embodiments or figures may be performed sequentially or in parallel (e.g., in a parallel processor or multithreaded environment), and may include:

S201, acquiring a plurality of accounts related to target behaviors and target behavior execution information corresponding to each account.

In the embodiment of the present application, a plurality of accounts are obtained for a target behavior, and the target behavior is not limited, and is related to a specific scenario. For example, to tag a financial account, the target activity may be a transfer activity. To tag an account in a medical setting, the target behavior may be hospitalization, inquiry, etc. The target behavior execution information refers to information related to the target behavior of the corresponding account, for example, time for implementing the target behavior, frequency for implementing the target behavior, etc., which is not specifically limited in the embodiment of the present application.

S202, extracting preset behavior co-occurrence conditions according to target behavior execution information corresponding to each account, generating a social network diagram according to an extraction result, wherein the preset behavior is a target behavior meeting preset requirements, and the accounts corresponding to two nodes connected by edges in the social network diagram have the same preset behavior.

The social network diagram of the embodiment of the application is a topological diagram determined according to the behavior of the account, and in different scenes, the social network diagram can be generated through different account behaviors, but the social network diagram is common in that the topology of the social network diagram is built according to the co-occurrence condition of preset behaviors. The preset behavior is not limited in this embodiment, and may be, for example, a certain account behavior or a certain account behaviors within a preset time interval. For example, in a partner mining scenario of financial fraud, a social network diagram may be generated according to financial operation behaviors of an account in a preset time interval, where the financial operation behaviors may include a single behavior or multiple behaviors, and the multiple behaviors are exemplified by a large transfer behavior, a frequent withdrawal behavior, and the like, and of course, a definition manner of "large amount" and "frequent" may be set according to actual situations. In the preset time interval, if two accounts respectively carry out large transfer and frequent withdrawal, an edge exists between the nodes corresponding to the two accounts respectively, because the two accounts have the same preset behaviors, namely the financial operation behaviors are executed in the preset time interval.

For another example, in a medical fraudulent party mining scenario, a social networking graph may be generated from medical data of an account. The application needs to propose that the data used in the implementation process of the application are legal data, namely, the embodiment of the application uses the security data which are fully authorized by the relevant subject and have no privacy risk. In some scenarios, the data that needs to be used may be desensitized. The data desensitization refers to the deformation of data of certain sensitive information through a desensitization rule, so that the reliable protection of sensitive privacy data is realized. Where customer security data or some commercially sensitive data is involved, the data therein may be modified to effect data desensitization without violating relevant rules.

The embodiment of the application can be applied to a plurality of partner mining scenes, and the following is mainly explained by taking the partner mining of medical fraud as an example, and other scenes are based on the same inventive concept and are not repeated.

In a medical fraud group mining scenario, the preset behavior may include a medical participation behavior performed by a target object corresponding to the account during a preset time interval, where the medical participation behavior includes a visit behavior, a settlement behavior involving in medical treatment, or a hospitalization behavior. In the embodiment of the application, the target object may refer to a person, such as a doctor-seeking person, a doctor-seeking family member, and the like.

In one embodiment, medical data such as data of treatment, settlement, hospitalization and the like can be utilized to construct a relationship among target objects of common treatment (settlement, hospitalization and the like) within m minutes according to concentrated treatment (settlement, hospitalization and the like) behaviors among the target objects, so that a social network diagram can be constructed

Wherein (1)>

Representing a set of nodes, ε represents a set of edges, where m is a natural number greater than 1. In the social network diagram, nodes represent accounts corresponding to target objects, and edges represent relationships between the target objects.

S203, classifying the nodes of the social network graph to obtain a first type node carrying a label and a second type node not carrying the label, wherein the label represents whether a corresponding account has a target social behavior.

The embodiment of the application does not limit the target social behavior, and can be one social behavior related to the target behavior. Taking a financial scenario as an example, the target action may be a large transfer and the target social action may be financial fraud. Taking a medical scenario as an example, the target behavior may be hospitalization or centralized visit, and the target social behavior may be medical fraud.

Taking the above medical fraud partner mining as an example, only one part of nodes in the social network diagram are labeled, the other part of nodes are not labeled, the labeled nodes belong to the first class of nodes, and the nodes without labels belong to the second class of nodes, so that the labels of the second class of nodes need to be predicted. For a medical fraud scenario, the tag may be a medical fraud tag that characterizes whether the account is in medical fraud. Specifically, the label of the node in the social network diagram is Y E R ^n×c . Where R represents a real set, n represents the number of target objects, and c represents the number of categories of tags.

Representing a set of labeled first class nodes,/->

Representing the set formed by nodes of the second class without labels,/->

And->

And

the node labels corresponding to the two node sets are Y respectively _l And Y _u 。

S204, carrying out label prediction on the nodes according to the information of each node in the social network diagram, and obtaining label distribution corresponding to each node.

The embodiment of the application is not limited to the method for performing label prediction, for example, a graph neural network or a transducer may be used to perform label prediction, specifically, in these models, node characteristics of each node may be obtained by extracting own information of the node, and label distribution may be obtained by performing label prediction based on the node characteristics.

In a specific embodiment, feature extraction may be performed on information corresponding to each of the nodes in the social network graph, so as to obtain node features corresponding to each of the nodes; transmitting the node characteristics of the node and the node characteristics of the neighbor nodes of the node to each node in the social network graph to obtain the aggregation characteristics corresponding to the nodes; and carrying out label prediction according to aggregation characteristics corresponding to the nodes respectively to obtain label distribution corresponding to each node.

Specifically, the embodiment of the present application does not limit information of nodes used for label prediction, and naturally, does not limit node characteristics corresponding to each node. In some embodiments, the node characteristic is derived from at least one of the following information for the corresponding account: static attribute information and account behavior information, and specific contents of the static attribute information and the account behavior information are not limited in the embodiment of the application. Taking medical fraud group mining scenario as an example, the static attribute information may be identification number, name, gender, native, age, history, allergen, medical history, medical participation, medical account, insurance reimbursement, funds, family members seek medical attention, etc. The account behavior information may be a hospitalization behavior, a settlement behavior, a hospitalization behavior, and the like. The node characteristics can be obtained by extracting the characteristics of the information, and the embodiment of the application is not limited to the characteristic extraction method, can realize the characteristic extraction by using a characteristic extraction neural network in the related technology, and can realize the characteristic extraction by carrying out the customized quantization on the information.

For each node, the node may use v _i To express, the purpose of feature delivery is to deliver neighbor node v _j V _i Is passed to node v _i . At the node v _i Feature information transferred to node v _i When in use, the transformation matrix W can be utilized ^l ) Representation of nodes

Performing feature transformation to enhance the expression capacity and transmitting the expression capacity as a target node v _i Is a piece of information of (a). />

The feature transfer can be performed for a plurality of times for the initial node feature of the node, that is, each time the feature transfer can change the node feature (node representation) of each node, and the feature transfer is performed for a plurality of times in an iteration manner, so that the feature information of each node is further plump, and a feature transfer result is finally obtained. l represents the first feature pass. X is x _i For row i of node feature matrix X, node v is represented _i Is a feature of the initial feature of (a). Node->

For node v _i Is->

Representing node v _i A set of neighbor nodes in a layer i iteration. Transformation matrix W # - ^l ) The transformation matrix at the time of the layer i iteration is represented, and the transformation matrix may be a parameter in the account label prediction model, which is a machine implementing steps S204 to S205, and the training method of the account label prediction model is described in detail below, which is not explained in detail herein.

During feature delivery, the neighbor node v can be obtained by _j Is transferred to the target node v _i When using the transformation matrix W ^(l) Node characteristics for neighbor nodes

Performing feature transformation, then averaging all the neighbor node features, and transmitting the feature transformation to the target node v _i Is of (1)And (5) extinguishing. The feature transfer operation may be expressed as follows:

wherein, the liquid crystal display device comprises a liquid crystal display device,

representing the transformation matrix. Message (i→i) represents node v after passing through the transformation matrix _i Node characteristics of (v), i.e. node v _i To itself. />

Representing node v _i Number of neighbor nodes at layer i. Message (j→i) represents averaging the representations of all neighbor nodes, i.e. all neighbor nodes are passed on to the target node v _i Is a piece of information of (a).

The connection relationships between nodes in the social network graph may reflect the association between target objects, which may also have similar characteristics. The purpose of feature aggregation is to transmit node features among nodes according to the connection relation among the nodes, after feature transmission, the node features of each node are updated, the node features are enriched, various node features related to the nodes can be further aggregated through feature aggregation, the aggregation features corresponding to the nodes are obtained, and the characterization capability of the aggregation features of the nodes is further improved. Information can be communicated between related target objects and aggregated to obtain an efficient node representation (aggregate characteristics) that can be directly used to predict the label distribution of the nodes. The feature aggregation operation may be expressed as follows:

representing node characteristics output after l times of characteristic transfer, message (·) representing operation of node characteristic transfer, and Message (i→i) representing node v _i Feature delivery to itself, message (j→i) representing neighbor node v _j To node v _i Feature transfer is performed. Aggregate (. Cndot.) represents the operation of node feature aggregation.

In the embodiment of the application, the target node may refer to any node in the social network diagram, and the feature aggregation is to aggregate the information of the target node and the information transferred by the neighbor node to obtain the node representation of the target node. A common aggregate operation is a splice operation. The feature aggregation operation may be expressed as follows:

Aggregate(Message(i→i),Message(j→i))＝Message(i→i)||Message(j→i)

the method comprises the steps of performing a splicing operation, and realizing feature aggregation through node features of a splicing target node and node features of neighbor nodes. In addition to the concatenation operation, a mean operation, a summation operation, a pooling operation, and the like can be used as the feature aggregation operation, which is not limited in the embodiment of the present application.

The embodiment of the application is not limited to the method of performing label prediction by using the aggregate features, for example, prediction can be completed by a fitting method or a pre-trained neural network, and the prediction can also be implemented by the account label prediction model.

S205, predicting the label of each second class node according to the labels carried by the first class nodes and label distribution corresponding to the nodes.

The labels carried by the first type nodes are known and accurate labels, and the distribution of each label is a predicted label, so that the accuracy is slightly lower, the known labels are used for influencing the distribution of each label, the accuracy of the distribution of each label can be improved, an error correction effect is achieved, and the error correction result is based on the error correction resultPredicting labels of the second class of nodes may improve prediction accuracy. Specifically, the label distribution set may be determined according to the label distribution corresponding to each of the above nodes. The label distribution set is composed of label distributions corresponding to the labels respectively. In one embodiment, the aggregated node expressions may be normalized to obtain label distribution corresponding to each node, so as to obtain the target node v _i For example, its corresponding tag distribution

I rows of the tag distribution matrix H are probability distributions of the nodes belonging to each type of tags, and the sum of probabilities of all tags of a node is 1.

Then, generating a tag information map according to the tag distribution set, wherein the tag information map comprises tag distribution in the tag distribution set and has the same topology as the social network map; and carrying out label propagation on the label information graph according to labels carried by the first class nodes to obtain labels of the second class nodes.

After the label prediction result of the second class node is obtained, each node in the social network diagram is provided with a label for representing whether the node has the target social behavior, so that the group partner with the group target social behavior can be mined on the premise. Specifically, community detection may be performed according to labels of the nodes in the social network diagram to obtain at least one cluster, where each cluster includes at least two nodes; and determining a target community according to the clusters and labels corresponding to each node in the clusters, wherein the target community represents an account cluster with a group target social behavior.

Taking the foregoing medical fraud group mining scenario as an example, the labels of each node characterize whether the node is medical fraudulent, i.e., each node is labeled with a "medical fraudulent" or not, that is, whether the account corresponding to each node is known as medical fraudulent. Based on the above, community detection can be performed according to labels of the nodes in the social network diagram to obtain at least one cluster, wherein each cluster comprises at least two nodes; and determining a target community according to the clusters and the labels corresponding to the nodes in the clusters. The target community can be considered as the community corresponding to the medical fraud group, so that the community discovery or the group mining is predicted to be completed based on the account label. The community detection algorithm is not limited, and community discovery algorithms based on modularity maximization, K-Clique and the like can be used for performing community division on nodes in a social network diagram, so that medical fraud groups are obtained. K-Clique is a community discovery algorithm, which is a well-known technique for identifying communities in a network. It is based on the insight that communities consist of K-Cliques, which refers to a complete subgraph with K vertices. Two communities (cliques) are considered to be adjacent if they share k-1 nodes. The K-Cliques community is composed of all adjacent K-Cliques.

In one example, in particular, a tag refers to an account being a normal or fraudulent user. After prediction, only knowing whether a single account has fraudulent activity, later community discovery is to cluster accounts, and when there are many fraudulent accounts within a community or class, the community can be considered as a fraudulent partner. The community division is carried out by utilizing a community detection algorithm aiming at the social network diagram. The basis can be understood as: the community division is to expect that the connection tightness of objects in the same community is higher than that of objects among different communities. When there are fraudulent accounts in a community, the likelihood that the community is a fraudulent party is high.

Referring to FIG. 3, a schematic diagram of partner mining through account prediction is shown. First, a relevant account is acquired. Specifically, a plurality of accounts related to the target behavior and target behavior execution information corresponding to each account are obtained. And then, after the accounts are extracted based on the co-occurrence condition of preset behaviors, constructing a social network diagram. And classifying the nodes of the social network graph to obtain a first class node carrying a label and a second class node not carrying the label, wherein the label characterizes whether the corresponding account has the target social behavior. According to the node information of each node in the social network diagram, after feature transfer and feature aggregation, the aggregate features of the node, namely node representation, can be obtained, and the node representation is fitted into label distribution of the node. And obtaining a label distribution set according to the label distribution of each node, wherein the label distribution set is used for generating a label information graph, and predicting the labels of the second type nodes without the labels by carrying out label propagation in the label information graph according to the labels of the first type nodes with the labels. And carrying out community discovery according to labels corresponding to the nodes respectively, thereby obtaining a result of the group mining.

In the following, referring to fig. 4, a flowchart of a training method of an account tag prediction model is shown, however, the structure of the model is not limited in this embodiment, the structure of the model may be understood as a neural network, the specific structure of the neural network is not limited, a related technology may be adopted, and only the operation required to be executed in the embodiment of the present application may be executed, and various neural networks described in the present application may be used to construct the structure of the model. The method comprises the following steps:

s401, acquiring a sample social network diagram, wherein the sample social network diagram comprises first type sample nodes carrying labels and second type sample nodes not carrying labels.

The meanings of the sample social network graph, the first type sample node and the second type sample node without the label are compared with those of the social network graph, the first type node and the second type node without the label in the foregoing, and are not described herein.

S402, executing the following operations based on a preset neural network: obtaining label distribution corresponding to each sample node by executing feature transfer and label prediction based on feature transfer results; performing label screening operation on labels carried by the first type of sample nodes to determine a third type of sample nodes and a fourth type of sample nodes, wherein the third type of sample nodes are sample nodes carrying the screened labels, and the fourth type of sample nodes belong to the first type of sample nodes and do not belong to the third type of sample nodes; obtaining a sample label distribution set based on labels carried by the third type of sample nodes and label distribution corresponding to each sample node respectively; predicting the prediction label of each fourth type of sample node according to the labels carried by the third type of sample nodes and the sample label distribution set.

And performing label screening operation on labels carried by the first type of sample nodes to determine a third type of sample nodes and a fourth type of sample nodes, wherein the third type of sample nodes are sample nodes carrying the screened labels, and the fourth type of sample nodes are sample nodes belonging to the first type of sample nodes and not belonging to the third type of sample nodes. The label screening method is not limited, for example, random screening may be performed, or the first type sample node with the account identification tail number of singular number is determined to be the third type sample node, and other first type sample nodes are determined to be the fourth type sample node. Specifically, a sample label distribution set may be determined according to label distributions corresponding to the sample nodes; and updating the sample label distribution set based on the labels carried by the third class of sample nodes. And replacing the label distribution of the corresponding label by using the label carried by the third type of sample node, so as to update the sample label distribution set, and predicting based on the updated sample label distribution set.

Specifically, labels carried by the third class of sample nodes in the sample label distribution set may be kept unchanged, and label propagation may be performed in a sample label information graph based on labels carried by the third class of sample nodes to obtain a predicted label of each fourth class of sample nodes, where the sample label information graph is obtained according to the sample label distribution set and has the same topology as the sample social network graph.

The basic idea of label propagation is to predict labels of unlabeled nodes with labels of labeled nodes based on node labels and relationships between nodes. The label propagation process is as follows: each node updates its own label according to the label of the adjacent node, and the larger the similarity between the label and the adjacent node is, the larger the influence weight of the adjacent node on the label is, so that the labels of the similar nodes tend to be consistent. During the label propagation process, the labels of marked nodes are kept unchanged, and the labels are propagated to unmarked nodes.

Y ^(k+1) ＝αA'Y ^(k) +(1-α)Y ⁽⁰⁾

representing the node label distribution output after k label passes. />

Is a label distribution matrix after label screening. Alpha E [0,1 ]]Is a weight parameter used to balance the weights of the updated tag and the original tag. />

Represents the adjacency matrix after normalization, an

Representing that a adjacency matrix with node self-connections is added, and +.>

Representing the identity matrix. />

Representing a degree matrix, and->

Representing node v _i Degree of (1)/(2)>

Besides the characteristics of the node itself and the labels, other parameters belong to the parameters of the preset neural network and can be optimized in the training process.

S403, according to the difference between the labels carried by the fourth type sample nodes and the predictive labels of the fourth type sample nodes, adjusting parameters of the preset neural network to obtain the account label predictive model.

The parameters can be adjusted based on a gradient descent method in the embodiment of the application. The gradient descent method is a method which is frequently used in the field of machine learning and deep learning for performing network parameter adjustment and is used for performing first-order optimization adjustment on network parameters in a gradient descent mode. The gradient descent method in the embodiment of the application can guide the parameter to adjust towards the direction of reducing the difference. And stopping parameter adjustment when the adjustment times reach a preset times threshold or when the difference is smaller than a preset difference threshold, so as to obtain the account label prediction model. Of course, embodiments of the present application are not limited to the above-described method of measuring the difference, and for example, a cross entropy loss measurement may be used.

Specifically, after label propagation, a label distribution matrix of all nodes is obtained, and a supervision label is extracted from the label distribution matrix

Corresponding node->

Predicted tag distribution Y 'of (2)' _l The following loss function can be constructed:

/>

is the true label of the marked node, +.>

Is a predictive label for the marked node. Finally, after the model is trained, the labels of the unlabeled nodes can be predicted.

In the above steps, the labels of the third type of sample nodes participate in the prediction process, and the labels of the fourth type of sample nodes participate in the reverse adjustment process, that is, serve as the supervision information, so steps S401-S403 are essentially a training method based on the supervision information, and of course, the preset neural network model can be built with reference to the neural network supporting the semi-supervision or the supervision training in the related art.

Since some nodes are in the embodiment of the application

With tag information, i.e. the tags of nodes of the first type are denoted Y _l Y can be used _l The corresponding tag distribution in the tag distribution set (tag distribution matrix H) is updated. In this way, label propagation can be performed by using the relationship among nodes, the labels of marked nodes and the label distribution of unmarked nodes, and the labels of unmarked nodes can be predicted.

However, if the node is marked

Tag Y of (2) _l On the one hand, the label propagation stage is to be participated in for predicting labels of unlabeled nodes. On the other hand, the model is trained as supervision information in the model training stage, so that the obtained label information can be fully utilized to improve the prediction accuracy, but leakage of the label information can be caused, the performance of the model is affected, and the improvement effect is reduced. To prevent the problem of label leakage during model training, the marked nodes are +.>

Tag Y of (2) _l Tag screening and screening are carried outSelecting a part of nodes +.>

The label of this part of the nodes is denoted +.>

Only tag +.>

For the tag propagation phase, i.e. with the tag +.>

Substitution of the corresponding node in the tag distribution matrix H>

And keeps the label distribution of other nodes unchanged. After tag screening, the remaining other part of nodes is +. >

The label of this part of the nodes is denoted +.>

Only tag +.>

As supervision information. Different label information is used in the label spreading stage and the feedback adjusting stage, so that the label leakage problem can be effectively prevented, and the model can be trained.

Referring to fig. 5, the embodiment of the present application further provides a flowchart of a training method, which is very similar to the training method described above, except that no tag screening is performed. The method comprises the following steps:

s501, acquiring a sample social network diagram, wherein the sample social network diagram comprises a first type sample node and a second type sample node, and the first type sample node and the second type sample node carry labels.

S502, executing the following operations based on a preset neural network: obtaining label distribution corresponding to each sample node by executing feature transfer and label prediction based on feature transfer results; obtaining a sample label distribution set based on labels carried by the first type sample nodes and label distribution corresponding to each sample node respectively; predicting the prediction label of each sample node of the second type according to the labels carried by the sample nodes of the first type and the sample label distribution set.

Determining a sample label distribution set according to label distribution corresponding to each sample node; and updating the sample label distribution set based on the labels carried by the first type of sample nodes. And maintaining the labels carried by the first type nodes in the sample label distribution set unchanged, and carrying out label propagation in a sample label information graph based on the labels carried by the first type nodes to obtain labels of the second type sample nodes, wherein the sample label information graph is obtained according to the sample label distribution set and has the same topology as the sample social network graph.

S503, according to the difference between the labels carried by the second type sample nodes and the prediction labels of the second type sample nodes, adjusting parameters of the preset neural network to obtain the account label prediction model.

Steps S501 to S503 are similar to the execution of steps S401 to S403, and thus will not be described in detail. Compared with the steps S401-S403, the method does not perform label screening, so that there may be a certain leakage of label information, thereby affecting the model training effect, but compared with the related art, the method still achieves the purpose of modeling according to the known labels and the known node characteristics, and compared with the related art, the prediction accuracy can be improved.

Referring to fig. 6, a schematic diagram of an implementation scheme of a training phase is shown, where the training phase is composed of three parts, i.e., feature aggregation, label propagation, and model training. First, a sample social network diagram is constructed, accounts are represented as nodes, and relationships generated by actions between the accounts are represented as edges connecting the nodes. In the training phase, first, a sample social network graph is input into the graph neural network, sample node features between sample nodes are passed for feature aggregation, and sample node representations are learned. And then mapping the sample node representation into label distribution of sample nodes, performing label screening on labels of part of sample nodes, and performing label propagation according to the relation among the sample nodes to obtain a final node label. And finally, constructing a neural network of the loss function training graph for predicting labels of unlabeled nodes, and finally training to obtain an account label prediction model. The embodiment of the present application does not limit the graph neural network, which is a specific form of the preset neural network in an example. Of course, the structure and parameters of the neural network in the embodiments of the present application may be changed, for example: the number of layers of the network, the dimensions represented by the nodes of the graph neural network, etc. The graph neural network can be constructed based on an encoder, and of course, the graph convolution network in the encoder can be replaced by a graph annotation network or other graph neural network. According to the embodiment of the application, the cluster partner diffusion mining can be realized only by the fact that a small number of labels for representing whether fraud information exists in an original sample, more fraud cluster partners are found, the time for manually marking data can be effectively reduced, and the training process of the model is quickened.

Referring to fig. 7, a schematic diagram of an implementation scheme of an application phase is shown, where the application phase is composed of three parts, namely feature aggregation, tag propagation and community discovery. Predicting labels corresponding to unlabeled nodes in the social network graph by using the trained account label prediction model, and carrying out community discovery on the social network graph to obtain at least one cluster, so that more fraudulent parties can be mined based on the cluster. The embodiment of the application can improve the detection performance of the partner diffusion excavation. Moreover, the account label prediction model may obtain node representations of the nodes that may be considered a comprehensive representation of the node information, and thus, further tasks, such as account portrayal tasks, fraudulent account detection tasks, fraudulent group discovery tasks, etc., may be performed based on the node representations.

According to the method, the relationship modeling of each account is conducted according to the behaviors of each account, so that a social network diagram is obtained. And carrying out label prediction on each node of the social network graph to obtain label distribution corresponding to each node. Based on the nodes with known labels in the social network diagram and label distribution corresponding to each node, the prediction labels corresponding to the nodes without labels can be predicted in a label propagation mode. In the method, the node information of each node and the label in the node with the known label are utilized in the process of obtaining the prediction label, and the information of the node and the information of the known label are effectively modeled, so that the accuracy of label prediction is improved. That is, when label prediction is performed, the characteristic information and the label information of each node in the social network diagram are effectively utilized, and the accuracy of label prediction of the nodes without the label information is improved. The node representation of each node can be learned by carrying out feature transfer and feature aggregation on the node features of each node in the social network diagram, and the node representation is mapped into label distribution corresponding to each node respectively. Based on the nodes with known labels in the social network diagram and label distribution corresponding to each node, the prediction labels corresponding to the nodes without labels can be predicted in a label propagation mode. By utilizing the node characteristics of each node and the labels in the nodes with known labels, the two parts of information of the characteristics and the labels are effectively modeled, so that the accuracy of label prediction is improved. The foregoing prediction process may be implemented based on the account label prediction model obtained by training in the embodiment of the present application, and after labels of each node are obtained by prediction, community division may be performed on accounts respectively represented by each node by using a community discovery algorithm, so as to obtain at least one cluster, where the cluster in the embodiment of the present application may represent a result of partner mining.

In a specific medical scenario, in the medical insurance anti-fraud partner diffusion mining task, partner data labeling becomes difficult due to the concealment of partner cases, and manual labeling of partner data requires a lot of time and cost, so that only a small number of accounts are usually labeled as to whether to be fraudulent or not, and a large number of accounts are not labeled. In order to mine more rogue partners, it is necessary to predict other tags and discover and mine rogue partners with a small number of tags known, and detect more rogue partners. According to the method and the device for predicting the account label, the account label prediction model is obtained through training, other labels can be predicted under the condition that a small number of labels are known, and then the group partner diffusion mining is conducted based on the prediction result.

Referring to fig. 8, a block diagram of an account tag prediction apparatus according to the present embodiment is shown, where the apparatus includes:

an information obtaining module 801, configured to obtain a plurality of accounts related to a target behavior, and target behavior execution information corresponding to each of the accounts;

the tag prediction module 802 is configured to extract preset behavior co-occurrence conditions according to target behavior execution information corresponding to each account, generate a social network graph according to an extraction result, where the preset behavior is a target behavior that meets a preset requirement, and accounts corresponding to two nodes connected by an edge in the social network graph respectively have the same preset behavior; node classification is carried out on the social network graph to obtain a first type node carrying a label and a second type node not carrying the label, and the label represents whether a corresponding account has target social behavior or not; carrying out label prediction on the nodes according to the information of each node in the social network diagram to obtain label distribution corresponding to each node; predicting the label of each second class node according to the labels carried by the first class nodes and the label distribution corresponding to the nodes.

In one embodiment, the tag prediction module 802 is configured to perform the following operations:

extracting characteristics of information corresponding to each node in the social network diagram to obtain node characteristics corresponding to each node;

transmitting the node characteristics of the node and the node characteristics of the neighbor nodes of the node to each node in the social network graph to obtain the aggregation characteristics corresponding to the nodes;

and carrying out label prediction according to aggregation characteristics corresponding to the nodes respectively to obtain label distribution corresponding to each node.

determining a label distribution set according to label distribution corresponding to each node;

generating a tag information map according to the tag distribution set, wherein the tag information map comprises tag distribution in the tag distribution set and has the same topology as the social network map;

and carrying out label propagation on the label information graph according to labels carried by the first class nodes to obtain labels of the second class nodes.

In one embodiment, the apparatus further includes a community detection module configured to perform the following operations:

performing community detection according to labels of the nodes in the social network graph to obtain at least one cluster, wherein each cluster comprises at least two nodes;

and determining a target community according to the clusters and labels corresponding to each node in the clusters, wherein the target community represents an account cluster with a group target social behavior.

In one embodiment, the apparatus is implemented based on an account tag prediction model that is trained by:

acquiring a sample social network diagram, wherein the sample social network diagram comprises a first type sample node carrying a label and a second type sample node not carrying the label;

based on a preset neural network, the following operations are executed: obtaining label distribution corresponding to each sample node by executing feature transfer and label prediction based on feature transfer results; performing label screening operation on labels carried by the first type of sample nodes to determine a third type of sample nodes and a fourth type of sample nodes, wherein the third type of sample nodes are sample nodes carrying the screened labels, and the fourth type of sample nodes belong to the first type of sample nodes and do not belong to the third type of sample nodes; obtaining a sample label distribution set based on labels carried by the third type of sample nodes and label distribution corresponding to each sample node respectively; predicting the prediction label of each fourth type of sample node according to the labels carried by the third type of sample nodes and the sample label distribution set;

And adjusting parameters of the preset neural network according to the difference between the labels carried by the fourth type sample nodes and the predictive labels of the fourth type sample nodes to obtain the account label predictive model.

In an embodiment, the obtaining a sample tag distribution set based on the tag carried by the third type of sample node and the tag distribution corresponding to each sample node includes:

determining a sample label distribution set according to label distribution corresponding to each sample node;

and updating the sample label distribution set based on the labels carried by the third class of sample nodes.

In one embodiment, predicting the prediction label of each of the fourth class of sample nodes according to the labels carried by the third class of sample nodes and the sample label distribution set includes:

and maintaining labels carried by the third type of sample nodes in the sample label distribution set unchanged, and carrying out label propagation in a sample label information graph based on the labels carried by the third type of sample nodes to obtain a predicted label of each fourth type of sample node, wherein the sample label information graph is obtained according to the sample label distribution set and has the same topology as the sample social network graph.

acquiring a sample social network diagram, wherein the sample social network diagram comprises a first type sample node and a second type sample node, and the first type sample node and the second type sample node carry labels;

based on a preset neural network, the following operations are executed: obtaining label distribution corresponding to each sample node by executing feature transfer and label prediction based on feature transfer results; obtaining a sample label distribution set based on labels carried by the first type sample nodes and label distribution corresponding to each sample node respectively; predicting the prediction label of each second type sample node according to the labels carried by the first type sample nodes and the sample label distribution set;

and adjusting parameters of the preset neural network according to the difference between the labels carried by the second type sample nodes and the predictive labels of the second type sample nodes to obtain the account label predictive model.

In one embodiment, the obtaining a sample tag distribution set based on the tag carried by the first type sample node and the tag distribution corresponding to each sample node includes;

and updating the sample label distribution set based on the labels carried by the first type of sample nodes.

In one embodiment, the predicting the prediction label of each sample node of the second type according to the labels carried by each sample node of the first type and the sample label distribution set further includes:

and maintaining the labels carried by the first type nodes in the sample label distribution set unchanged, and carrying out label propagation in a sample label information graph based on the labels carried by the first type nodes to obtain labels of the second type sample nodes, wherein the sample label information graph is obtained according to the sample label distribution set and has the same topology as the sample social network graph.

The device part and the method embodiment in the embodiment of the present application are based on the same inventive concept, and are not described herein in detail.

Further, fig. 9 shows a schematic hardware structure of an apparatus for implementing the method provided by the embodiment of the present application, where the apparatus may participate in forming or including the device or the system provided by the embodiment of the present application. As shown in fig. 9, the apparatus 10 may include one or more processors 102 (shown as 102a, 102b, … …,102 n) that may include, but are not limited to, a processing device such as a microprocessor MCU or a programmable logic device FPGA, a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, the method may further include: a display, an input/output interface (I/O interface), a Universal Serial Bus (USB) port (which may be included as one of the ports of the I/O interface), a network interface, a power supply, and/or a camera. It will be appreciated by those skilled in the art that the configuration shown in fig. 9 is merely illustrative and is not intended to limit the configuration of the electronic device. For example, the device 10 may also include more or fewer components than shown in fig. 9, or have a different configuration than shown in fig. 9.

It should be noted that the one or more processors 102 and/or other data processing circuits described above may be referred to generally herein as "data processing circuits. The data processing circuit may be embodied in whole or in part in software, hardware, firmware, or any other combination. Further, the data processing circuitry may be a single stand-alone processing module, or incorporated in whole or in part into any of the other elements in the device 10 (or mobile device). As referred to in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of the path of the variable resistor termination to interface).

The memory 104 may be used to store software programs and modules of application software, such as program instructions/data storage devices corresponding to the methods described in the embodiments of the present application, and the processor 102 executes the software programs and modules stored in the memory 104 to perform various functional applications and data processing, that is, implement an account tag prediction method described above. Memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, memory 104 may further include memory located remotely from processor 102, which may be connected to device 10 via a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.

The transmission means 106 is arranged to receive or transmit data via a network. Specific examples of the network described above may include a wireless network provided by a communications provider of device 10. In one example, the transmission device 106 includes a network adapter (NetworkInterfaceController, NIC) that can connect to other network devices through a base station to communicate with the internet. In one example, the transmission device 106 may be a radio frequency (RadioFrequency, RF) module for communicating wirelessly with the internet.

The display may be, for example, a touch screen type Liquid Crystal Display (LCD) that may enable a user to exchange with a user interface of the device 10 (or mobile device).

It should be noted that: the foregoing sequence of the embodiments of the present application is only for describing, and does not represent the advantages and disadvantages of the embodiments. And the foregoing description has been directed to specific embodiments of this application. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims can be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

All embodiments in the embodiments of the present application are described in a progressive manner, and identical and similar parts of all embodiments are mutually referred to, and each embodiment focuses on the differences from other embodiments. In particular, for the device and server embodiments, since they are substantially similar to the method embodiments, the description is relatively simple, and reference is made to the description of portions of the method embodiments where relevant.

It will be understood by those skilled in the art that all or part of the steps for implementing the above embodiments may be implemented by hardware, or may be implemented by a program for instructing relevant hardware, where the program may be stored in a computer readable storage medium, and the storage medium may be a read-only memory, a magnetic disk or an optical disk, etc.

The instructions in the storage medium may perform an account label prediction method, the method comprising:

extracting preset behavior co-occurrence conditions according to target behavior execution information corresponding to each account, generating a social network diagram according to an extraction result, wherein the preset behavior is a target behavior meeting preset requirements, and the accounts corresponding to two nodes connected by edges in the social network diagram have the same preset behavior;

carrying out label prediction on the nodes according to the information of each node in the social network diagram to obtain label distribution corresponding to each node;

In one embodiment, the performing label prediction on the nodes according to the information of each node in the social network graph to obtain label distribution corresponding to each node includes:

In one embodiment, predicting the label of each of the second class nodes according to the label carried by each of the first class nodes and the label distribution corresponding to each of the nodes includes:

In one embodiment, the method further comprises:

In one embodiment, the method is implemented based on an account tag prediction model that is trained by:

The foregoing description of the preferred embodiments is merely exemplary in nature and is not intended to limit the embodiments of the present application, but is intended to cover any modifications, equivalents, alternatives, and improvements falling within the spirit and principles of the embodiments of the present application.

Claims

1. A method for predicting account labels, the method comprising:

2. The method of claim 1, wherein the performing label prediction on the nodes according to the information of each node in the social network graph to obtain label distribution corresponding to each node includes:

carrying out feature transfer on node features of the node and node features of neighbor nodes of the node to each node in the social network graph to obtain an aggregation feature corresponding to the node;

3. The method according to claim 1 or 2, wherein predicting the label of each node of the second class according to the label carried by each node of the first class and the label distribution corresponding to each node, respectively, comprises:

determining a label distribution set according to label distribution corresponding to each node respectively;

generating a tag information graph according to the tag distribution set, wherein the tag information graph comprises tag distribution in the tag distribution set and has the same topology as the social network graph;

4. The method according to claim 1, wherein the method further comprises:

5. The method of claim 3, wherein the method is performed based on an account tag prediction model trained by:

based on a preset neural network, the following operations are executed: obtaining label distribution corresponding to each sample node by executing feature transfer and label prediction based on feature transfer results; performing label screening operation on labels carried by the first type of sample nodes to determine a third type of sample nodes and a fourth type of sample nodes, wherein the third type of sample nodes are sample nodes carrying the screened labels, and the fourth type of sample nodes belong to the first type of sample nodes and do not belong to the third type of sample nodes; obtaining a sample label distribution set based on labels carried by the third type of sample nodes and label distribution corresponding to each sample node respectively; predicting the prediction label of each fourth type of sample node according to the labels carried by each third type of sample node and the sample label distribution set;

6. The method of claim 5, wherein the obtaining a sample label distribution set based on labels carried by the third class of sample nodes and label distributions respectively corresponding to the sample nodes includes:

7. The method of claim 5, wherein predicting the predicted labels for each of the fourth class of sample nodes based on the labels carried by each of the third class of sample nodes and the sample label distribution set, comprises:

and keeping labels carried by the third type of sample nodes in the sample label distribution set unchanged, and carrying out label propagation in a sample label information graph based on the labels carried by the third type of sample nodes to obtain a predicted label of each fourth type of sample nodes, wherein the sample label information graph is obtained according to the sample label distribution set and has the same topology as the sample social network graph.

8. The method of claim 3, wherein the method is performed based on an account tag prediction model trained by:

based on a preset neural network, the following operations are executed: obtaining label distribution corresponding to each sample node by executing feature transfer and label prediction based on feature transfer results; obtaining a sample label distribution set based on labels carried by the first type sample nodes and label distribution corresponding to each sample node respectively; predicting the prediction label of each second type sample node according to the labels carried by each first type sample node and the sample label distribution set;

and adjusting parameters of the preset neural network according to the difference between the labels carried by the second type sample nodes and the prediction labels of the second type sample nodes to obtain the account label prediction model.

9. The method of claim 8, wherein the obtaining a sample label distribution set based on labels carried by the first type of sample nodes and label distributions respectively corresponding to the sample nodes includes;

10. The method of claim 8, wherein predicting the predicted labels for each of the second type of sample nodes based on the labels carried by each of the first type of sample nodes and the sample label distribution set, further comprises:

the labels carried by the first type nodes in the sample label distribution set are kept unchanged, label propagation is carried out in a sample label information graph based on the labels carried by the first type nodes, the labels of the second type sample nodes are obtained, and the sample label information graph is obtained according to the sample label distribution set and has the same topology as the sample social network graph.

11. An account label prediction apparatus, the apparatus comprising:

12. A computer readable storage medium having stored therein at least one instruction or at least one program loaded and executed by a processor to implement an account label prediction method as claimed in any one of claims 1 to 10.

13. An electronic device comprising at least one processor, and a memory communicatively coupled to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the at least one processor implementing an account tag prediction method as claimed in any one of claims 1 to 10 by executing the instructions stored by the memory.

14. A computer program product comprising a computer program or instructions which when executed by a processor implements an account label prediction method as claimed in any one of claims 1 to 10.